Structural model of the gas vesicle protein GvpA and analysis of GvpA mutants in vivo

Authors


E-mail pfeifer@bio.tu-darmstadt.de; Tel. (+49) 6151 16 2957; Fax (+49) 6151 16 2956.

Summary

Gas vesicles are gas-filled protein structures increasing the buoyancy of cells. The gas vesicle envelope is mainly constituted by the 8 kDa protein GvpA forming a wall with a water excluding inner surface. A structure of GvpA is not available; recent solid-state NMR results suggest a coil-α-β-β-α-coil fold. We obtained a first structural model of GvpA by high-performance de novo modelling. Attenuated total reflection (ATR)-Fourier transform infrared spectroscopy (FTIR) supported this structure. A dimer of GvpA was derived that could explain the formation of the protein monolayer in the gas vesicle wall. The hydrophobic inner surface is mainly constituted by anti-parallel β-strands. The proposed structure allows the pinpointing of contact sites that were mutated and tested for the ability to form gas vesicles in haloarchaea. Mutations in α-helix I and α-helix II, but also in the β-turn affected the gas vesicle formation, whereas other alterations had no effect. All mutants supported the structural features deduced from the model. The proposed GvpA dimers allow the formation of a monolayer protein wall, also consistent with protease treatments of isolated gas vesicles.

Introduction

Gas vesicles are gas-filled proteinaceous structures found in aquatic bacteria such as cyanobacteria. These gas-filled structures increase the buoyancy of the cells and enable them to float towards the surface of natural water bodies (Walsby, 1994). Haloarchaea also produce these structures that are spindle- or cylinder-shaped up to 1 µm in length and 200 nm in diameter. The gas vesicle wall is solely constituted of proteins with the 8 kDa protein GvpA as the dominant species. The inner surface of the gas vesicle wall is hydrophobic, whereas the outer surface is more hydrophilic. Attached here is the second protein, GvpC, that stabilizes the gas vesicle wall. Gas vesicles are passively filled by diffusion with gas molecules dissolved in the cytoplasm. Water molecules might enter the structure, but the hydrophobic and curved inner surface prevents condensation (Walsby, 1994). Collapsed gas vesicles indicate a rib-shaped structure with 5 nm ribs running perpendicular to the long axis that are presumably formed by GvpA (Stoeckenius and Kunau, 1968; Blaurock and Walsby, 1976; Blaurock and Wober, 1976). The amino acid sequence of GvpA is highly conserved (Fig. 1), whereas the sequences of GvpC are more divergent. Fourteen gas vesicle protein (gvp) genes are involved in the formation of gas vesicles in the haloarchaeon Halobacterium salinarum. These genes are arranged in two oppositely oriented clusters, gvpACNO and gvpDEFGHIJKLM, located in the vac region (Englert et al., 1992a). Eight of the gvp genes (gvpA, F, G, J, K, L, M, O) are essential for gas vesicle formation as demonstrated by a deletion of single gvp genes in Haloferax volcanii p-vac transformants (Offner et al., 2000). The additional Gvp proteins either serve as gene regulators (GvpD, GvpE) (Hofacker et al., 2004; Scheuch et al., 2008) or are present in tiny amounts in gas vesicle preparations (GvpF, G, J, L, M) when prepared by flotation in 5% NaCl (Shukla and DasSarma, 2004).

Figure 1.

Alignment of GvpA sequences and prediction of α-helices and β-sheets. The 51-amino-acid conserved core region is indicated by a bar and the sites accessible to peptidases (Trp: trypsin, GluC) are marked by arrows (Belenky et al., 2004). pGvpA and cGvpA derive from Hbt. salinarum, mcGvpA from Hfx. mediterranei, nvGvpA from Halorubrum vacuolatum and hqGvpA from Haloquadratum walsbyi. Ms, Methanosarcina. The model structure is given in the last line with (H) depicting α-helical and (E) depicting β-sheet regions. For comparison, the PSIPRED and NMR results for GvpA of Anabaena are added (Sivertsen et al., 2010). The differences between PSIPRED and NMR results are marked by small letters (h = α-helical, e = β-sheet, additionally found in NMR, but not in PSIPRED).

A structure of GvpA is not yet available, but recent solid-state NMR results suggest a coil-α-β-β-α-coil fold (Sivertsen et al., 2010). Solution NMR studies are not feasible, since GvpA monomers have a high tendency to aggregate and dissolve only in 80% formic acid. Dialysis to remove the formic acid causes amorphous precipitates of GvpA rather than a refolded protein structure or even a reassembly of the gas vesicles (Belenky et al., 2004). Fourier transform infrared spectroscopy (FTIR) spectra obtained with gas vesicles indicate antiparallel β-sheets; X-ray diffraction and atomic force microscopy suggest that the β-strands of GvpA are tilted in the ribs at an angle of 54° to the axis of the rib (Blaurock and Walsby, 1976; McMaster et al., 1996). Determination of the complete amino acid sequence of GvpA as well as matrix-assisted laser desorption/ionization time of flight (MALDI-TOF) mass spectrometry indicates that GvpA is not post-translationally modified (Hayes et al., 1986; Belenky et al., 2004).

The peptide bonds of GvpA exposed in the gas vesicle structure have been determined by proteolytic cleavage using trypsin and other proteases and gas vesicles of Anabaena flos-aquae and Hbt. salinarum (Belenky et al., 2004). A 51-amino-acid (aa) highly conserved inner segment of GvpA is inaccessible to proteolytic cleavage in either case (Fig. 1). In the case of gas vesicles from Hbt. salinarum, only the C-terminal K60–I61 bond of GvpA is accessible to trypsin, whereas other possible trypsin sites located in the conserved 51 aa inner segment are not affected. Cleavage at K60 results in collapsed gas vesicles (Belenky et al., 2004). None of the possible GluC endopeptidase sites (D–X or E–X) within the conserved segment of GvpA is accessible, only the ones in the C-terminal portion (Fig. 1) implying that the C-terminal portion of GvpA is located at the outer surface of the gas vesicles. Solid-state NMR studies performed with gas vesicles of A. flos-aquae suggest non-equivalent GvpA subunits in the gas vesicle structure due to a small folding variation in the alternating GvpA subunits (Sivertsen et al., 2009). The model derived implies that the β-sheet portion of GvpA achieves a hydrophobic surface, and complementary charges and aromatic–aromatic interactions are present at the subunit interfaces.

In this report we present a structural model of GvpA obtained by de novo modelling. The GvpA sequence derived from the haloarchaeon Haloferax mediterranei is almost identical to the pGvpA sequence of Hbt. salinarum (Fig. 1). Haloarchaea offer a genetic transformation system enabling the study of pGvpA mutants and their effect on gas vesicle formation by complementing transformants harbouring the ΔA construct carrying all p-gvp genes of p-vac except p-gvpA of Hbt. salinarum (Offner et al., 2000). Point mutations and small deletions were introduced in pGvpA at positions that are – according to the model – important for the structure. The pGvpA mutants were tested for their ability to form gas vesicles in Hfx. volcaniiΔA+Amut transformants harbouring in addition to ΔA construct Amut containing the p-gvpA mutant gene.

Results

Structural model of GvpA

Template-based modelling and secondary structure prediction.  The amino acid sequence of mcGvpA derived from Hfx. mediterranei was used to obtain a structural model in silico. No alignments with significant prediction scores could be found using 3DJury. The FUGUE server delivered a single marginally significant result. While tertiary structure prediction using template-based modelling proved difficult, all secondary structure prediction methods resulted in similar secondary structure elements (Fig. S1). Helical regions were predicted between L9–V16 and V48–I58, while a β-strand was predicted for V22–V32, both in good agreement with recent results from solid-state NMR (Sivertsen et al., 2010). PSIPRED's secondary structure prediction seems the most plausible as it permits a pairing of two equal-length β-strands with another β-strand from E35 to E47; again in correspondence with solid-state NMR (Sivertsen et al., 2010) (see Fig. 1). PHYRE also failed to detect homology to known proteins, but its secondary structure prediction element (using sspro) coincides with the PSIPRED secondary structure prediction.

In continuation to the 3DJury results FUGUE was directly used as an alignment server. A marginally significant result (Z-score: −3.95) was obtained for the PDB-template 2JOI. A model was built using Modeller, 2JOI and the FUGUE alignment. In contrast to the previous hypothesis of two paired strands, the model shows four paired strands stabilized by two helices. Since a large fraction of the amino acids in this model remained unstructured, this approach was not pursued further.

De novo modelling.  Since no models of promising quality emerged from 3DJury, separate FUGUE, I-Tasser and SAM-T08 modelling attempts, we resorted to a de novo prediction using the predicted secondary structure elements as constraints. In total 53 000 main-chain models were generated using ROSETTA, which generated 1850 centroid models after clustering. The best energy structure (Fig. 2) has a PFF02 energy of −163 kcal mol−1 and is separated by a gap of about 8 kcal mol−1 to the next structure of different topology (Figs S2 and S3).

Figure 2.

Ribbon illustration of the best energy model of a GvpA monomer. A single central β-strand (red) is found comprising of two bonded β-strands from positions V23 to L31 and E35 to V43. Furthermore there is a long helix (blue) from V48 to T67 and a smaller helix from amino acids L9 to K19. The accessible surface area is coloured using the Eisenberg-Schwarz hydrophobicity scale encoded in the intensities of green and includes the two lysine residues that were found in solid-state NMR (Sivertsen et al., 2010) to be solvent-accessible.

The resulting model features two bonded β-sheet regions at V23–L31 and E35–V43, each of which is 9 aa long (Fig. 1). In addition there is one long α-helix (V48–T67) that is arranged almost in the same direction as the β-sheets but faces the other surface. These regions agree with the secondary structure predictions we used to generate the original models. Another α-helix is found from amino acid L9 to amino acid K19. Eleven hydrogen bonds are located between the two strands of the β-sheet. The whole population of models generated by ROSETTA has a large fraction of secondary structure elements (Fig. S4). Refining these models with PFF02 thus selected the best energy model among many competitors with similar secondary structure.

Results of the molecular dynamics.  The RMSD over the five independently simulated trajectories is small for a de novo predicted protein (Fig. S5 shows the RMSD versus the simulated time). The Spearman correlations of the root-mean square fluctuations to the elastic network model (see Experimental procedures) are around 0.68, which is significant. The root mean square fluctuations between the two high-salt scenarios differ only slightly with a high degree of correlation as shown in Fig. S6.

Individual dihedral angles in the secondary structure elements showed overall small deviations from the ones in the predicted structure (data not shown). In particular, differences between the two salt regimes were minor, indicating the structure to have only a small dynamical susceptibility to ion concentrations. While the side-chain orientations changed during the simulation frequently, the dihedral angles showed for most, if not all, sites to be in range of the monomer stable configurations. These results suggest the validity of the predicted structure.

Dimerization.  Among the 622 000 decoys the best dimerization energy of −128.917 kcal mol−1, with the second best structure showing an energy of −128.700 kcal mol−1. However, the two structures have an RMSD of only 0.09 Å– rendering them equivalent (Fig. 3). Clearly a close to symmetric dimerization occurs. The ‘inner’ concave surface of the dimer shows a remarkable hydrophobicity. Due to this observation and the small bending angle within the dimer, we speculate this to be the inner surface of the gas vesicle. Such a structural mode would be in perfect agreement with results from solid-state NMR (Sivertsen et al., 2010).

Figure 3.

Best dimer structure obtained from docking. The dimer is represented as two of the monomers depicted in Fig. 2.
A. Surface of the GvpA dimer at the convex outer surface of the gas vesicle, left: cartoon picture of secondary structure elements, right: volume rendering. The N- and C-terminus within a monomer are colour-coded.
B. Concave, gas containing cavity facing surface of GvpA dimers. Two dimers each are arranged to form a rib; the width of each rib is 4.3 nm. Grey = hydrophobic residues, blue = non-hydrophobic residues, orange = mutated residues.

Results of contact importance scoring.  As stated above the root mean square fluctuations of the MD simulation of the monomer are in good agreement with the fluctuations computed within the elastic network model. Based on this observation we judge on the relevance of individual contacts within the monomer structure by elastic network model sensitivity analysis. The most important interactions are the ones connecting the loop regions to stable secondary structure elements. This indicates that the secondary structure elements are stabilized not only by themselves, but also by multiple interactions between each other, as the biggest changes occur if intrinsically flexible loops are ‘released’ by missing contacts to the stabilizing α-helices and β-sheets (Fig. S7).

Infrared spectroscopy of GvpA

Infrared spectra are indicative of the protein secondary structure composition in the Amide I+II region (1700–1500 cm−1) (Jackson and Mantsch, 1995; Goormaghtigh et al., 2006). Attenuated total reflection (ATR)-FTIR spectroscopy was therefore used to complement the secondary structure predictions based on sequence analysis.

Purified gas vesicles were freed from GvpC to obtain pure GvpA. For this purpose, gas vesicle preparations were washed at least four times by centrifugation accelerated flotation in water. Western analysis indicated the lack of GvpC (Fig. 4A) and other minor Gvps (F, G, L, M) were also not detectable (data not shown). The spectra of sonicated GV were compared with that of the proteasome, a soluble protein complex of known structure, containing 36% α-helix and 27% β-strands according to Löwe et al., (1995). The spectra in Fig. 5A show prominent peaks that are attributed to the contribution of α-helix/random coil (1655–1652 cm−1), β-sheet (1638 cm−1), β-structure/intermolecular aggregates (1627 cm−1) in the Amide I region, and of predominantly helix (1548 cm−1) and a mixture of α- and β-structure (1536 cm−1) in the Amide II region. H–D exchange leads to band shifts in the Amide I region that allows us to distinguish between α-helix and random structures more clearly. The components at 1650 cm−1 (proteasome) and 1649 cm−1 (GvpA) visible in the spectra of deuterated proteins and enhanced after Fourier self-deconvolution (FSD) (Fig. 5B) originate from α-helices while random or loop structures that are expected to create absorptions around 1645 cm−1 play a minor role. The spectrum of deuterated GvpA revealed a peak at 1636 cm−1 that is characteristic for antiparallel β-strands as well as the spectral band at 1692 cm−1 (Jackson and Mantsch, 1995); the component at 1664 cm−1 is indicative of β-turns. The shifted major component at 1623 cm−1 originates from intermolecular β-sheet-like interactions that have not necessarily to be built from β-strands alone (Jackson and Mantsch, 1995; Zandomeneghi et al., 2004). Extended β-strand interactions as observed in antiparallel β-barrels of bacterial outer membrane proteins lead to prominent peaks close to 1630 cm−1 (Zeth et al., 1998). It is likely that regular GvpA–GvpA contacts that stabilize the gas vesicle wall are responsible for the signal at 1623 cm−1. The proteasome is a soluble protein and does not form extended aggregates; the characteristic peak is thus missing (Fig. 5B).

Figure 4.

Western analysis to determine the presence of GvpC in gas vesicle preparations (A) and GvpA in cell lysates (B).
A. Gas vesicles prepared by three- to four-times centrifugation accelerated flotation in 5% NaCl (GV) or water (GV3, S3, GV4, S4) were separated by SDS-PAGE and analysed for the presence of GvpC using an antiserum raised against GvpC (Englert and Pfeifer, 1993). A cell lysate of Hfx. volcanii producing GvpC (lysate) and isolated GvpC protein (GvpC) were used as controls. GV, gas vesicles; S, supernatant solution.
B. Presence of GvpA in lysates of ΔA+Amut transformants indicated on top. The antiserum was raised against gas vesicles of Hfx. mediterranei and exclusively detects GvpA (Englert et al., 1992b). The location of the GvpA monomer is marked by an arrow.

Figure 5.

Infrared spectra of GvpA (red curves) and the proteasome (black curves) for comparison. (A) ATR-FTIR spectra in the Amide I+II region before and (B) in the Amide I region after H–D exchange. The peak positions (cm−1) of original spectra (solid lines) and spectra enhanced by Fourier self-deconvolution (dotted lines) are marked.

Quantitative estimates of the secondary structure composition are hardly possible because of the aggregation of GvpA and the strong effect in the IR spectra. Since sonication only disrupts the gas vesicles into large fragments and GvpA monomers are insoluble (Belenky et al., 2004) reliable CD measurements were impossible. The fragments scatter UV light (data not shown) and impair the CD spectrum in a way that cannot be compensated for easily. As a result, FTIR spectroscopy suggests that the major structure elements of GvpA are α-helix and antiparallel β-strands with a quantitative preference for helical structures according to the shape of the Amide II band (Goormaghtigh et al., 2006).

GvpA mutants and gas vesicle formation in vivo

The in silico model of GvpA predicted two α-helices (helix I, L9–K19 and helix II, V48–T67), and two β-strands of 9 aa length (V23–L31 and E35–V43; Fig. 6). For an initial in vivo analysis of the GvpA structure eight point mutants and three deletion mutants of GvpA were produced and their effect on gas vesicle formation was studied in Hfx. volcaniiΔA+Amut transformants where construct ΔA (containing all p-gvp genes except p-gvpA) is complemented by a mutated p-gvpA gene. The presence of gas vesicles is easily detectable by visual inspection of the colony phenotype on agar plates, which become turbid when cells contain gas vesicles. In contrast, colonies formed by cells lacking gas vesicles are red and transparent. The Hfx. volcanii transformants were also analysed by phase-contrast (light) microscopy (PCM) and transmission electron microscopy (TEM). Gas vesicles isolated from colonies grown for 6–7 weeks on agar plates were used for size and shape determinations.

Figure 6.

Mutations in the GvpA sequence (A) and location of the amino acids altered in the model structure of GvpA (B).
A. Sequence of GvpA and α-β-β-α model structure below. The various point mutations are indicated in bold and the C-terminal deletions by arrows above the sequence. The helical and β-sheet regions are marked.
B. Positions of mutations in the structure of GvpA. The side-chains of the mutated amino acids are indicated.

The point mutations in GvpA were S6A (serine present at position 6 substituted by alanine), S7A, R15A, R15K, I34M, E35A, F51Y and K60L (Fig. 6). The location of these mutations in the structure of GvpA is shown in Fig. 6B. In addition, three deletion variants of GvpA were constructed lacking 5, 7 or 11 aa at the C-terminus (Fig. 6). In each case, the presence of the mutations in gvpA was confirmed by DNA sequence analysis. The mutant proteins were tested in the respective Hfx. volcaniiΔA+Amut transformants, and the presence of GvpA or GvpAmut was determined by Western analysis (Fig. 4B). Aggregates of GvpA were seen near the top of the gel, whereas the 8 kDa monomer was barely visible. The control transformant ΔA+A carrying the p-gvpA wild-type gene formed pink and turbid colonies, and cells inspected by TEM showed the typical spindle-shaped gas vesicles after 7 days of growth (Fig. 7). Gas vesicles isolated 6 weeks later were significantly larger with an average length of 850 nm (Fig. 7, bottom; Table S1).

Figure 7.

Effect of deletions in the C-terminus of GvpA. Wild-type transformant ΔA+pA is shown in comparison with the ΔA+Δ5, ΔA+Δ7 and ΔA+Δ11 transformants. Colonies on agar plates after 5–6 weeks are shown in the top row, TEM of the cells in the middle and isolated gas vesicles analysed by TEM in the bottom row. A phase-contrast micrograph of ΔA+Δ11 transformants is given on the right.

Deletions at the C-terminus of GvpA.  The GvpA deletion variants Δ5 and Δ7 lacking 5 or 7 aa at the C-terminus formed gas vesicles in ΔA+Amut transformants as seen by the pink and turbid appearance of colonies on agar plates (Fig. 7). Cells grown in liquid culture and left standing for a few days on the bench floated to the surface (Fig. S8). The presence of gas vesicles was also confirmed by TEM (Fig. 7), suggesting that the C-terminal 7 aa of GvpA are not important for the formation of the gas vesicle wall. Isolated gas vesicles were cylinder-shaped and of similar length as determined for the wild type (Table S1). In contrast, colonies of ΔA+Δ11 were transparent and the cells were unable to float in liquid media suggesting the lack of gas vesicles (Fig. 7 and S8). However, cells derived from colonies grown for 7 weeks contained tiny light refractile bodies (Fig. 7). Rarely, very small gas vesicles of 100 nm lengths were found. The Δ11 mutant lacks the relatively variable C-terminal sequence of GvpA including the last 2 aa of helix II. Thus, the C-terminal portion of helix II is important for the formation of long and stable gas vesicles. In contrast, the unstructured C-terminal region of GvpA is dispensable.

Amino acid substitutions in GvpA.  The two serine residues S6 and S7 are part of the central conserved 51 aa region of GvpA (Fig. 1). Both amino acids are close to helix I in our GvpA model (Fig. 6B). Serine 7 is turned to the outside of GvpA and is possibly able to mediate contact to an adjacent GvpA monomer. Both ΔA+S6A and ΔA+S7A transformants produced gas vesicles (Fig. 8). Colonies on agar plates were turbid, and when liquid cultures were left standing on the bench the cells floated at the surface (Fig. S8). Younger cells inspected by TEM showed small groups of spindle-shaped gas vesicles, and longer gas vesicles similar to wild type were found after 7 weeks (Fig. 8). A minor difference was that gas vesicles of ΔA+S7A had a tendency to clump. This feature is visible in the TEM micrographs where the gas vesicles attached to each other throughout the surface (Fig. 8).

Figure 8.

Effect of point mutations in GvpA. Top row, colonies of agar plates after 5–6 weeks; middle, TEM of transformant cells; bottom row, isolated gas vesicles of the respective mutants. Mutants ΔA+K60L were lysed directly on the grid. Mutants R15A and R15K do not contain gas vesicles.

The arginine residue R15 present in the centre of α-helix I was altered to alanine (R15A) or lysine (R15K). The positive charge of R15 is pointing outside and might be important for binding another GvpA molecule to form the gas vesicle wall. Both ΔA+Amut transformants lacked gas vesicles (Figs 8 and S8). The colonies were transparent, and cells of both cultures sink to the bottom of the tube (Fig. S8). Cells inspected by TEM lacked gas vesicles indicating that R15 is essential for the formation of the gas vesicle wall. The lack of gas vesicles in R15K implied that the guanidino group of arginine was required since the positively charged lysine cannot restore the function of the protein in the formation of gas vesicles.

The most exposed structure found in the GvpA monomer is the predicted β-turn formed by the three amino acids V32–G33–I34 (Fig. 6B). All three amino acids are hydrophobic and should be involved in binding of an adjacent GvpA in the gas vesicle wall. Close to this loop is the charged E35, the first amino acid of β-strand II. Isoleucine 34 was altered to methionine (adding the first sulphur containing amino acid to GvpA), and the adjacent E35 was altered to alanine (Fig. 6A). Colonies of both ΔA+Amut transformants were orange and turbid (suggesting the presence of gas vesicles), but both were unable to float (Figs 8 and S8). The analysis by TEM indicated that ΔA+I34A transformants contained very long and thin gas vesicles (Figs 8 and S9). When isolated after 7 weeks, gas vesicles were on average 1.48 ± 0.53 µm long and 0.12 ± 0.08 µm wide and easily attached at the long side (Fig. 8; Table S1). Each gas vesicle maintained a constant diameter throughout the entire length of the structure suggesting that the enlargement in diameter was disturbed. The ΔA+E35A transformants contained gas vesicles of similar length as found with wild-type GvpA, but these were with 0.16 µm smaller compared with the 0.25 µm average diameter determined for the wild type (Figs 8 and S9). These differences suggested that these two amino acids present in the tip of GvpA have an impact on the gas vesicle shape, but do not prevent the formation of the GvpA envelope.

Two mutations were introduced in helix II, F51Y and K60L (Fig. 6A). Both alterations changed the amino acid sequence of the haloarchaeal GvpA to the respective amino acid found in the cyanobacterial GvpA (Fig. 1). The side-chain of K60 points outside of the GvpA whereas F51 is pointing inside (Fig. 6B). Both GvpA mutants yielded gas vesicles in transformants, but these were altered in size (K60L) or strength (F51Y). The ΔA+F51Y transformant produced gas vesicles similar to wild type, but cells grown in liquid culture sink to the bottom of the tube (Figs 8 and S8). In contrast, colonies of ΔA+K60L were transparent (Fig. 8). Inspection of cells by TEM yielded very small gas vesicles after 7 days of growth that could not be isolated by flotation but were detected by direct lysis of the cells on the grid used for TEM. This result implied that lysine 60 is of importance for gas vesicle formation in haloarchaea.

Discussion

The major gas vesicle structural protein GvpA forms a water-excluding but gas-permeable protein wall with a hydrophobic inner and hydrophilic outer surface. Neither lipids nor carbohydrates are present. A structure of GvpA was obtained by in silico modelling of the 76 aa GvpA sequence and site-directed mutagenesis was performed to test the mutant proteins for their ability to form gas vesicles in vivo.

GvpA contains significant α-helical regions

The proposed GvpA structure contains two α-helices (I and II) separated by two 9-aa-long anti-parallel β-strands of 3.67 nm length. Overall the structure contains 47% helical, 25% β-sheet and 28% of unknown structure. The antiparallel β-strands presumably form the inner surface of the gas vesicle wall. Nine out of 10 residues are hydrophobic and point towards the gas-facing surface of the wall, whereas six out of eight amino acids, pointing to the other side of the β-sheet, are hydrophilic/charged. Helix I is also located at the gas-facing surface and forms the broader back of the triangular-shaped GvpA monomer (Fig. 3). This region is opposite to the ‘tip’ of GvpA formed by the hydrophobic β-turn. For cyanobacterial gas vesicles it has been speculated that GvpA consists of β-strands only (Walsby, 1994). However, the presence of a significant amount of α-helical structure in GvpA of haloarchaea was confirmed by FTIR spectroscopy with isolated gas vesicles washed of GvpC. Immunological studies showed that these gas vesicle preparations contained GvpA but lacked GvpC and other Gvp proteins (Fig. 4A and data not shown). The analysis proposed a significant amount of α-helical structures, but also two-stranded antiparallel β-sheets that are connected. This larger β-sheet structure could be formed by aggregation of the GvpA dimers as suggested below.

Our structural results and their assignment to secondary structure elements agrees very well with the NMR results obtained for cyanobacterial GvpA of Anabaena (Sivertsen et al., 2010) (see Fig. 1). The only differences occur in the C-terminal region, where the haloarchaeal and cyanobacterial GvpA sequences substantially differ.

GvpA mutations confirm the formation of the proposed GvpA monolayer

The GvpA dimer contains two antiparallel GvpA monomers of triangular shape that are connected by contacts between half of the antiparallel β-sheet region of monomer 1 and monomer 2 (Fig. 3). The relatively large extension (tip) formed by the second half of the anti-parallel β-strands and the β-turn should contact the next GvpA dimer located in the adjacent rib, with one monomer contacting the rib on top and the other one the rib below as indicated in Fig. 3B. It is striking how perfectly the dimers fit to each other when a single layer of GvpA is formed. The orientation of the antiparallel β-sheet relative to the axis of the rib corresponds well with the 55° angle shown by X-ray diffraction studies of cyanobacterial gas vesicles (Blaurock and Walsby, 1976). Such periodicities also are observed by atomic force microscopy (McMaster et al., 1996). The unit cell of the GvpA monomer model measures 4.3 nm (across the rib without the tip region) × 2.2 (along the rib defined by the dimer structure) × 2.1 nm (wall thickness) (Fig. 3). These lengths were measured from the most distant atoms along the respective axes, effectively neglecting the unknown separation of unit cells. This is close to the dimensions of the repeating unit cell (4.57 × 1.15 × 1.95 nm) known from fibre X-ray crystallography using gas vesicles from the cyanobacterium A. flos-aquae (Blaurock and Walsby, 1976). It should be noted that the length of the β-sheet (3.67 nm) is significantly shorter compared with the width of the rib (4.3 nm) and thus constitutes only part of it. However, the tip of an adjacent GvpA monomer (constituted by antiparallel β-sheet + β-turn) interacts with the β-sheet of a second GvpA monomer, which might result in the strong interactions between two adjacent ribs.

The structural model of GvpA deduced here ascribes putative contact sites of GvpA. The major contact sites besides the β-turn at the tip are helix I (near the bottom) and helix II (alongside the dimer). The mutations performed in GvpA confirmed the importance of these regions. The alterations are indicated in orange in Fig. 3B. The alteration I34M in the β-turn resulted in extremely long and thin cylindrical gas vesicles, and the alteration E35A close by resulted in cylinder-shaped gas vesicles. Thus, the single amino acid alterations in GvpA have a large impact on the gas vesicle shape. Smaller changes of the gas vesicle shape were already observed when the pGvpA sequence was altered to cGvpA or mcGvpA of Hfx. mediterranei, where ΔA+mcA resulted in narrower gas vesicles compared with ΔA+pA (Beard et al., 2002).

Two amino acid substitutions in helix II also affected the shape. This helix presumably is alongside in contact with the adjacent GvpA molecule in the same rib. Lysine K60 is pointing outside, whereas F51 is pointing inside the GvpA molecule (Fig. 6B). The K60L mutation in GvpA resulted in tiny gas vesicles that did not enlarge but often stayed attached to each other in a small group, suggesting that the enlargement of the gas vesicle structure was disturbed. The mutation F51Y yielded gas vesicles of similar size as found in wild type. The importance of helix II was also demonstrated by mutant Δ11 since the deletion of the last 2 aa inhibited gas vesicle formation in ΔA+Δ11 transformants. In contrast, ΔA+Δ5 and ΔA+Δ7 transformants still contained gas vesicles, suggesting that this variable C-terminal portion of GvpA is not so important. The C-terminal region is highly divergent in GvpA proteins of different haloarchaeal species, and is completely missing in GvpA of Haloquadratum walsbyi (Fig. 1). This region is accessible to endopeptidase GluC (leaving gas vesicles intact), whereas the only site accessible to trypsin is K60–I61 and leads to the collapse of gas vesicles upon cleavage (Belenky et al., 2004).

The most interesting GvpA mutants carried the alteration of arginine 15 in the centre of helix I (R15A and R15K). Both mutations resulted in the lack of gas vesicles suggesting that the guanidino group plays an important role in the aggregation of GvpA. Our structural model of GvpA implies that R15 and other charged amino acids of helix I might bind a second GvpA dimer of the adjacent rib. The charged amino acid residues of helix I are highly conserved between haloarchaeal and cyanobacterial GvpA proteins (Fig. 1). It is possible that these charged amino acids are in very close contact to the amino acids with opposite charges of helix I of the other GvpA dimer. Further mutations in helix I are under way to test their importance for gas vesicle formation.

The structural model of GvpA allows for the first time to derive a model on the aggregation of GvpA in the gas vesicle wall. The data are consistent with a hydrophobic gas-facing surface constituted by β-sheets and a more hydrophilic outer surface. The proposed contact sites connect adjacent GvpA monomers within a rib (helix II) or to the adjacent ribs at the bottom (helix I) or top (β-turn) and allow the formation of this protein wall. The in vivo data on the GvpA mutants support the major structural features deduced from the model of GvpA. Further GvpA mutants will help to refine our model on the formation of the gas vesicle wall.

Experimental procedures

Template-based modelling

Templates were obtained by interrogating the 3DJury (Ginalski et al., 2003) Meta-Server. Among the servers queried were PSIPRED (Bryson et al., 2005) and two SAM-T02 (Karplus et al., 2003) secondary structure prediction servers. Furthermore the PHYRE fold recognition server (Kelley and Sternberg, 2009) was used to identify the secondary structure in similar sequences. Finally a separate alignment using FUGUE (Shi et al., 2001) was created, which was translated into a three-dimensional model using the Modeller program. The GvpA protein was also modelled directly by the I-Tasser (Zhang, 2008) and SAM-T08 (Katzman et al., 2008) web-based threading servers.

De novo structure prediction

The de novo structure prediction scheme constitutes a model generation step and a refinement step. In the model generation step we use ROSETTA (Rohl et al., 2004) to generate a set of decoys, which are clustered using MaxCluster in order to avoid rescoring the same model multiple times. Afterwards the decoys are ranked using the all-atom free-energy force field PFF02 as a scoring function. Since models generated with one method rarely score perfectly in another, independent force field, each of the models must be relaxed in a short simulation to a close-by local minimum in the PFF02 force field.

Fragment-based modelling

Fragments were generated using the Robetta server (Bonneau et al., 2002). We assembled these fragments locally into main-chain models using the ROSETTA 2.3 software suite and the default ab initio protocol. Subsequently a short idealization run was carried out, which added the missing side-chain atoms and removed clashes in the ROSETTA force field. The resulting ensemble of models was clustered using MaxCluster with a 3DJury (MaxSub) (Siew et al., 2000) score threshold of 0.55 and a 3DJury pair threshold of 20.

All-atom refinement

POEM (protein optimization using energy methods) is an all-atom free-energy protein simulation package implementing the free-energy model PFF02 (Verma and Wenzel, 2009). PFF02 is an all-atom free-energy potential comprising five terms modelling electrostatics, angle-dependent hydrogen bonding, Lennard-Jones and solvent interactions (accessible surface area-based implicit solvent model), as well as an empirical backbone torsion potential. This force field was demonstrated to select near-native decoys for all 32 monomeric proteins (without cofactors) from the ROSETTA decoy set (Tsai et al., 2003) and used to fold a set of 24 proteins with helical, sheet and mixed secondary-structure in de novo simulations (Verma and Wenzel, 2009). In this study we applied a refinement protocol similar to that used in the CASP7/8 evaluations (Gopal et al., 2009) using POEM@HOME, which implements POEM in a distributed volunteer computing network enabling the screening of a large set of model candidates in parallel.

Each of the MaxCluster cluster centroids was relaxed in five independent runs per model, using a simulated annealing protocol with a starting/end temperature of 700 K/5 K, respectively, and a geometrical cooling protocol. Note that these temperatures are exclusively chosen for the computational method of simulated annealing and are not real temperatures. Each simulation comprised 1.5 million steps, in which one randomly selected dihedral angle of the model was changed by a maximum of five degrees. Backbone/side-chain dihedral angles were selected for rotation with a ratio of 70%/30% (Verma et al., 2006).

Molecular dynamics simulations

To judge the stability of the predicted monomer, we simulated the dynamics of this molecule with the NAMD molecular dynamics package (Phillips et al., 2005) for 20 ns at two different salt concentrations in reference to the halophilic organisms: (i) at 1 M KCl and (ii) at 1 M NaCl and 5 M KCl in the CHARMM force field (MacKerell et al., 2004). Trajectory analysis was performed by programs in the GROMACS software suite (Van der Spoel et al., 2005). We repeated the simulations five times for each salt concentration.

Molecular docking

The ROSETTA Protein–Protein–Docking protocol (Wang et al., 2007) was used and 622 000 decoys were scored for a potential GvpA heterodimer.

Structural importance of contacts.  We applied the Anisotropic Network Model (Atilgan et al., 2001; Hoffgaard et al., 2010) with uniform interaction of contacts of k = 3.14 kcal (mol Å2)−1 and peptide bond strength K = 82 kcal (mol Å2)−1 to the predicted structure of the GvpA monomer (Hamacher, 2008; Hamacher and McCammon, 2006). We recomputed the correlation matrix of amino acid movements in this framework for all scenarios with vanishing interaction strength for all of the predicted contacts individually. This constitutes a sensitivity analysis towards presence or absence of individual contacts. Computing the Frobenius norm of the covariance matrices of the ‘fully interacting’ structure to the matrices for all cases with vanishing interaction strengths gives a semi-quantitative understanding of the relevance each of the contacts has for the overall molecular dynamics and stability.

Mutagenesis of p-gvpA

The gvpA mutants were constructed in Escherichia coli using the wild-type gene present in construct pApD-WT containing the p-gvpA gene in pSK+ as template. The various gvpA mutants contained single or more nucleotide substitutions to achieve the desired exchange in pGvpA. The mutations were incorporated by polymerase chain reaction (PCR) using synthetic oligonucleotides carrying the mutations as primers (Table S2). The mutations were: S6A (serine present at position 6 substituted by alanine), S7A, R15A, R15K, I34M, E35A, F51Y and K60L. Deletion variants lacking the last 5, 7 or 11 amino acids at the C-terminus were constructed by introducing additional stop codons in the p-gvpA reading frame.

In the case of substitution mutants the two oligonucleotides used for PCR were head to head so that the synthesis of DNA resulted in a linear vector molecule with blunt ends. In case of the deletions, the resulting linear fragment lacked the desired nucleotides and carried the inserted stop codon at the end. The PCR products were treated with DpnI to hydrolyse the template. The DNA was purified, phosphorylated at the 5′ ends, ligated and E. coli was transformed with the resulting vector plasmids. Each mutation in p-gvpA was confirmed by DNA sequence analysis.

The respective p-gvpA mutant gene (including the native promoter) was transferred as XbaI–HindIII fragment to the haloarchaeal vector plasmid pMDS20 and the resulting Amut construct was used to transform Hfx. volcanii (Pfeifer and Ghahraman, 1993). Hfx. volcanii is easy to transform, grows faster than Hbt. salinarum and lacks any endogenous gvp genes thus providing a clean genetic background. ΔA transformants carrying all p-gvp genes except for p-gvpA (Offner et al., 2000) were complemented with the various Amut constructs and tested for gas vesicle formation. The presence of each construct was confirmed by Southern analysis. The presence of GvpA was confirmed by Western analysis using an antiserum raised against isolated gas vesicles of Hfx. mediterranei that detects GvpA (Englert et al., 1992b). Western analysis was performed as described (Pfeifer et al., 2001).

PCM and TEM

Gas vesicles were determined by the inspection of cells grown in liquid culture or on solid medium using PCM or TEM (Hechler and Pfeifer, 2009). The cells were observed in a Zeiss EM109 electron microscope equipped with a Gatan MultiScan 600 W camera. Electron microscopy was also carried out with purified gas vesicles isolated by accelerated flotation (Shukla and DasSarma, 2004). Cells taken from colonies grown on agar plates were lysed in 1 mM MgSO4•6H2O + 10 µg ml−1 DNase I for 2 h. Gas vesicles were purified by flotation overnight at 4°C, 1 h centrifugation at 60 g, and resuspension in 5% NaCl, 100 mM Tris-HCl, pH 7.2.

Preparation of gas vesicles and proteasome for FTIR

Gas vesicles were isolated from 1 l of cultures grown for 14 days and left standing for 7 days. Cells floating at the surface were collected and lysed by the addition of 1 mM MgSO4•6H2O + DNase I (final concentration 10 µg ml−1) for 2–3 h at 37°C. The lysate was overloaded with 5% NaCl, 100 mM Tris-HCl, pH 7.2 and centrifuged at 60 g, 4°C for 16 h. Gas vesicles were collected from the surface and again treated with 5% NaCl, 100 mM Tris-HCl, pH 7.2 for centrifugation accelerated flotation. Repeating this procedure with Tris-HCl-buffered water for at least four times resulted in gas vesicles lacking GvpC, whereas this protein remained at the gas vesicles after repeated flotation in 5% NaCl. The presence or absence of GvpC was determined by Western analysis using an antiserum raised against GvpC protein (Englert and Pfeifer, 1993). The presence of additional accessory Gvp proteins (GvpF, G, L, M) in both gas vesicle preparations was checked by Western analysis using the appropriate antisera. None of these Gvp proteins was detectable (data not shown). Gas vesicles resuspended in aqua dest. were exposed to ultrasonication (water bath at room temperature) for ≈ 5 s to disrupt intact vesicles for infrared spectroscopy. Proteasomes from Thermoplasma acidophilum were purified as recombinant protein expressed in E. coli according to Zwickl et al. (1992). The protein was finally dissolved in aqua dest.

FTIR spectroscopy

Fourier spectra were measured in the FTIR spectrophotometer Vertex 70 from BRUKER (Ettlingen, Germany) equipped with a TGS detector by means of the ATR technique using parallelogram-shaped germanium (Ge) crystals (2 × 45°; 25 × 10 × 3 mm) as internal reflection plates (Korth Kristalle, Altenholz, Germany). The Ge plates were cleaned with aqua dest., ethanol, toluol, and finally exposed in a plasma cleaner (Harrick, Ossining, NY) for 1 min. A thin film of 50–100 µg of purified protein (dissolved in aqua dest.) was dried under nitrogen stream on one side of the Ge crystal that was placed in a home-made gastight chamber afterwards. H–D exchange was performed by flushing D2O-saturated nitrogen gas through the chamber inside the spectrophotometer and monitored every 5–10 min until the kinetics of band shifts were insignificant (usually after > 60 min). Spectra were recorded before and after H–D exchange with a nominal resolution of 2 cm−1 in the double-sided, forward–backward mode, collecting 1024 scans per sample. Water vapour and CO2 contributions were corrected by using the atmospheric compensation of the OPUS software (version 6.5) from BRUKER. FSD of spectra was performed applying the parameter half width = 14.5 cm−1 and noise reduction = 0.45. Illustrations of spectral data were created by means of the IGOR PRO software system, version 6.12 (WaveMetrics, Lake Oswego, Oregon).

Acknowledgements

We thank Sarah Breuer for the kind gift of a proteasome sample. K.H. is grateful for financial support of junior faculty by the Fonds der Chemischen Industrie. T.S. and W.W. acknowledge support from the Baden-Württemberg Stiftung GmbH (HPC and Biofunctional Surfaces Programs). We also thank the volunteers at POEM@HOME for their support. F.P. is grateful for the continuous financial support by the Deutsche Forschungsgemeinschaft (DFG, PF 165/10). Molecular structures were rendered by VMD (Humphrey et al., 1996), using the SURF algorithms for surfaces (Varshney et al., 1994). Tony Walsby is thanked for many valuable suggestions to the manuscript.

Ancillary